TikTok and ByteDance run the most coding-heavy MLE loops in big tech. Nearly every 45-60 minute technical round has the same shape: a resume and project deep-dive, twenty minutes of ML theory, and a LeetCode problem in the last twenty, medium in screens and medium-to-hard later, with every round graded pass/fail. Dynamic programming (coin change, grid paths, LCS), grid graphs (islands, shortest bridge), binary search on real numbers, sliding windows and linked-list surgery dominate the reports, and a single untested edge case has ended loops the same day. ML coding from scratch is the second gate: self-attention with masking and multiple heads above all, then logistic regression with its gradient, MLP backprop in NumPy, inverse-transform sampling, layer norm, KL and cross-entropy. Theory centres on transformers and the KV cache, long-context tricks, optimizers, overfitting, the role of the bias term, trees versus forests and precision versus recall; Algorithm Engineer and LLM-track loops add PPO versus DPO versus GRPO, quantization and LLM parameter arithmetic. Where a design round exists it is recommendation or ads: candidate generation, multi-task ranking, re-ranking, cold start, position bias and the offline-online metric gap. Early-career candidates start with a HackerRank OA (CS MCQs plus a medium and a hard); an HR round usually closes the loop; interviews can run in Mandarin with China-based panels, and headcount can vanish mid-process, so keep other offers moving.
It depends on your starting point and the role. Deep-ML builds a paced, company-specific plan (2 weeks to 3 months) from a TikTok / ByteDance interview path, the right resume projects, and timed mock interviews, then tracks your readiness so you know when you're ready.